Flush of regular mcast messages could trigger entry into gather state if the
regular mcast message in the kernel buffers contained a foriegn message. This
should set orf_token_dicard to 1. Unfortunately if this happens, the following
ORF tokens in the kernel buffer that are flushed should be rejected. Since
the flush occurs after the check for orf_token_discard==1, the orf token is
not rejected.
Signed-off-by: Steven Dake <sdake@xxxxxxxxxx>
---
exec/totemsrp.c | 13 ++++++++++---
1 files changed, 10 insertions(+), 3 deletions(-)
diff --git a/exec/totemsrp.c b/exec/totemsrp.c
index 5a78962..9c7b01d 100644
--- a/exec/totemsrp.c
+++ b/exec/totemsrp.c
@@ -3451,9 +3451,6 @@ static int message_handler_orf_token (
"Time since last token %0.4f ms\n", ((float)tv_diff) / 1000000.0);
#endif
- if (instance->orf_token_discard) {
- return (0);
- }
#ifdef TEST_DROP_ORF_TOKEN_PERCENTAGE
if (random()%100 < TEST_DROP_ORF_TOKEN_PERCENTAGE) {
return (0);
@@ -3499,6 +3496,16 @@ static int message_handler_orf_token (
totemrrp_recv_flush (instance->totemrrp_context);
/*
+ * Discard orf tokens in GATHER/COMMIT after recv_flush operation.
+ *
+ * It is possible a foreign message caused a shift to the GATHER
+ * state.
+ */
+ if (instance->orf_token_discard) {
+ return (0);
+ }
+
+ /*
* Determine if we should hold (in reality drop) the token
*/
instance->my_token_held = 0;
--
1.7.7.3
_______________________________________________
discuss mailing list
discuss@xxxxxxxxxxxx
http://lists.corosync.org/mailman/listinfo/discuss
